Developer Thomas Ptacek Sets New Rule for Using LLMs as Copyeditors, Not Writing Assistants

Thomas Ptacek, a software engineer and prolific blogger, published a short guide on Simon Willison’s website on 17 September 2026, urging writers to treat large language models as copyeditors rather than as writing assistants. The post opens with a hard rule: "Rule Number One: You may not use a single word an LLM suggests to you." Ptacek explains that he uses LLMs for fact‑checking, spelling, grammar, and a quick thesaurus, but never for drafting prose.

Key points

• Ptacek’s rule forbids using any LLM‑suggested phrase.
• He employs LLMs for post‑editing tasks like fact‑checking and spell‑checking.
• A screenshot of his personal copyediting tool appears in the article.
• The post includes a prompt template to help readers build their own tool.
• The article is a link post on Simon Willison’s site, dated 17 September 2026.

What happened

Ptacek’s article was posted on Simon Willison’s blog on 17 September 2026, a platform known for its focus on web‑development and open‑source culture. In the post, he writes, "I think that as a form of intellectual personal protective equipment you should adopt the rule that any specific turn of phrase an LLM suggests is off limits." The piece ends with a call to sponsors, offering a curated email digest of LLM developments for $10 per month.

The backstory

Large language models have become mainstream tools for drafting and editing content. Many developers and content teams treat them as co‑writers, feeding prompts and accepting the output as final text. Ptacek’s stance emerges from a growing concern that LLMs can imprint generic phrasing and reduce authorial voice. By positioning the model as a background helper for mechanical tasks—fact‑checking, spell‑checking, and occasional thesaurus use—he seeks to preserve the human element in narrative construction.

Why the rule matters

The discipline Ptacek advocates forces writers to confront the model’s suggestions critically. By rejecting every LLM‑generated turn of phrase, authors keep control over diction, tone, and nuance. This approach also mitigates risks of inadvertent plagiarism or the spread of subtle hallucinations that models can introduce when they produce content verbatim from their training data.

Trade‑offs for teams and publishers

Adopting Ptacek’s rule slows the writing workflow. Teams that rely on LLMs for rapid drafting will need to allocate additional time for manual review and rewriting. However, the trade‑off is a higher quality, more authentic voice that can differentiate a brand in a crowded content market. For publishers, the risk of over‑reliance on AI‑generated prose—often perceived as bland or repetitive—reduces, potentially boosting reader engagement.

What builders and businesses should do

• Treat LLMs as post‑editing assistants, not as primary writers.
• Build a copyediting pipeline that includes fact‑checking, grammar, and spelling checks powered by LLMs.
• Create a custom prompt template for your team, inspired by Ptacek’s example, to standardize the review process.
• Train authors to critically evaluate LLM suggestions, rejecting any phrase that feels generic or off‑tone.
• Monitor content quality metrics before and after implementing the rule to quantify impact.

The bottom line

Ptacek’s rule pushes a shift toward disciplined AI use in content creation. Businesses that adopt a copyediting‑first approach can maintain authorial integrity while still reaping the speed benefits of LLMs for routine tasks. The next wave of AI‑driven tools will likely need to support this model‑as‑assistant paradigm, making it a key factor for future product roadmaps.